Re: 1teraflops cell processor possible?

> +---------------
> | My solution to that was an architectural requirement for a 'yield'
> | instruction to be called every (say) M instructions or N memory
> | references, whichever comes first, and to abort the process if it
> | failed to do so.  Dead easy to implement.
> +---------------
> 
> Ahhh... Sutherland's "Wheel of Reincarnation" strikes again...  ;-}  ;-}

Another example: the transputer. There, the "yield" instruction is a
a jump or any comms instruction. Although it was theoretically possible to
lock out other processes at the same priority by running "non-yielding" 
code, I've never heard of a case, and the compilers certainly didn't care 
- "normal" code contains enough of the above set of instructions...

And, of course, all the other benefits mentioned elsethread, in particular
fast context switch times, were realised on the transputer: After such 
instructions, the evaluation stack was documented as being unpredictable,
so that the (hardware) scheduler only had to save one word (the PC in a
slot of the workspace provided for that purpose).

        Jan